How reliably can species co-occurrence be inferred from occupancy models?

Interspecific interactions, such as predation, competition, and mutualism, are fundamental to the structure and function of any biotic community. Understanding these interactions provides valuable insights into community dynamics and informs conservation efforts. One common approach to studying these interactions is through co-occurrence patterns, which remain essential for understanding species relationships. Occupancy models have emerged as powerful tools to investigate these patterns, offering a flexible framework for ecological questions. Among these models, the co-occurrence framework developed by Rota et al. (2016) has seen widespread applications across various species and habitats.

In this study, we used simulations to evaluate the performance of the Rota et al. (2016) co-occurrence model under different study designs and species interaction scenarios. Specifically, we examined how the bias and precision of species occupancy estimates are influenced by the presence or absence of another species, varying the number of sampling units surveyed and the number of temporal replicates per unit. Additionally, we compared the co-occurrence model’s performance with a simpler model that does not account for species interactions to determine when statistically significant co-occurrence patterns emerge.

We observed that occupancy parameters become more precise as the intensity of species avoidance increased, and while the number of sampling sites significantly influenced precision, the number of temporal replicates did not. Notably, the co-occurrence model outperformed the non-interaction model only when species avoidance was strong and at least 150 sites were sampled. Based on these results, we recommend a minimum of 150 sampling sites when applying the Rota et al. (2016) model to evaluate co-occurrence between two potentially interacting species. Our study emphasizes the need for further development of methods capable of detecting interaction signals, even at low to moderate interaction levels, and in studies with fewer sampling sites.